Attachment Material
Figure 1~Figure 4 are some other results of our method.
Figure 1. (a): input image; (b): saliency map; (c): our result; (d): our result with pseudo-perspective transformation.
Figure 2. (a): input image; (b): our result.
Figure 3. (a): input image; (b): our result; (c): our result with pseudo-perspective transformation.
Figure 4. Another comparison with “Virtual Passepartouts”. (a): input image from Internet; (b): our result; (c): result from [RTMS12].
We use Photoshop to handle a real pseudo-3D painting (shown as Figure 4(a)) by adding certain black background regions to get a simple
2D image without 3D effect as is shown Figure 4(b). We take Figure 4(b) as an input image for our method and get the result as is shown Figure 4(c), our result is similar to the original famous real pseudo-3D painting indicating our method is feasible. But the real pseudo-3D painting usually adds shadows and artistic borders to show more stereoscopic effect, our future work will also start from this point to carry on certain corresponding researches to get more satisfactory results.
Figure 4. (a):a real pseudo-3D painting; (b): 2D image by photoshop; (c): our result with (b) as the input image.
However, since our method use visual saliency to determine the highlighted part of foreground object, some visual saliency region is not near the margin of the foreground object. As is shown in Figure 5(b), visual saliency regions focus on the center of the two flowers, leading that highlighted foreground region is not proper and cutting too much background region, getting an inharmonious picture
without strong pseudo-3D effect. Therefore, the type of images for our method has certain restriction.
Figure 5. A failure example. (a): input image; (b): saliency map ; (c): our result
